On Fri, 2014-08-15 at 07:57 -0400, Ron Kerry wrote:
On 8/15/14, 7:45 AM, Lars Marowsky-Bree wrote:
With ocfs2, it's fine. It still works though, even if we disable one
of the nodes. We have only old hardware so it would be pointless having ext4 nodes with only one active at a time: we may as well go back to our single file server. I'm not sure this holds. There's a penalty needed for syncing and locking. Have you benchmarked if a single node is really slower than two? Especially a write-heavy workload probably won't, and it'll go very much downhill if you have metadata-intensive jobs (e.g., creating/deleting/stating lots of files/directories).
The main reason for HA is, well, availability, not performance, thus going back to a single server is likely worse than this anyway.
True enough, but the performance of an active-active HA configuration with two or more nodes serving NFS or CIFS can be well in excess of what a single server is capable of doing. This all depends on the underlying disk hardware. Many time that hardware may be capable of far more bandwidth than a single server can drive by itself. In this active-active clustered environment the performance achievable by any single node will be less than what it can do on its own, but the combined performance of many nodes to the same shared-clustered disk will be able to reach the bandwidth capability of the underlying disk hardware.
If a single node can drive your disk hardware at peak bandwidth, than as Lars says you will get no performance benefit (in fact you get a performance degradation) from an active-active HA configuration ... just the availability benefit.
Hi. It's the availability which is more important here. We're never going to get performance anyway as we can't afford new hardware.
--
Ron Kerry rkerry@sgi.com Global Product Support - SGI Federal
-- To unsubscribe, e-mail: opensuse-ha+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-ha+owner@opensuse.org